OpenAI is releasing a new AI model, o1, that has been trained to answer more complex questions, similar to a human. It's being released alongside o1-mini, a smaller, cheaper version.
According to OpenAI, the new o1 is a step closer to human thinking. It is more efficient at writing code and solving multi-step problems than previous models. But it is also more expensive and slower to use than GPT-4o. OpenAI calls this version a “preview.”
users have access to both o1-preview and o1-mini now, while Enterprise and Edu users will have access early next week. OpenAI said it plans to offer access to o1-mini to all free ChatGPT.
See also: OpenAI partners with Condé Nast (publisher of Vogue)
Developer access to o1 is very expensive: In the API, o1-preview is $15 per 1 million input tokens or pieces of text parsed by the model and $60 per 1 million output tokens.
The training of o1 is fundamentally different from that of the company's other models. OpenAI o1 "has been trained using a completely new algorithm optimization and a new training dataset specifically tailored for it."
Previous GPT models learned to imitate patterns from their training data. o1 has been trained to solve problems on its own, using a technique known as reinforcement learning. In this method, the system is trained through rewards and punishments. It then uses a “chain of thought” to process questions, similar to how humans process problems step by step. OpenAI believes o1 will be more accurate.
The main thing that sets this new model apart from GPT-4o is its ability to handle complex problems much better than its predecessors, while also explaining its “reasoning.
See also: OpenAI wants to recognize students' AI-generated assignments
A company executive said that OpenAI tested o1 in qualifying exams for the International Mathematical Olympiad. GPT-4o solved only 13% of the problems correctly, while the new improved o1 was correct 83%.
In online programming competitions, known as Codeforces competitions, this new model reached the 89th percentile of participants.
However, o1 is not as efficient as GPT-4o at other things. It does not effectively share real-world knowledge. It also does not have the ability to browse the web or edit files and images.
The company believes it represents a whole new class of capabilities. o1 can reason and find answers to more complex problems. In fact, OpenAI has designed the interface in such a way that it shows the model’s reasoning steps. Those who have seen it say it closely mimics human thinking. OpenAI doesn’t believe in equating AI model thinking with human thinking. But the interface is intended to show how the model spends more time on deeper processing to solve problems.
The company is bringing reasoning capabilities to its models because it sees a future with autonomous systems, capable of making decisions and taking actions on your behalf.
See also: OpenAI: Launches voice functionality for ChatGPT
For now, however, o1's reasoning abilities are relatively slow and expensive.
The o1 model represents a significant advance in the field of artificial intelligence, particularly in its ability to process and interpret complex information. By applying innovative training and leveraging extensive data sets, OpenAI has enabled o1 to outperform previous models, approaching human reasoning abilities. This development opens up a wide range of potential applications, from advanced customer support to complex data analysis, thereby revolutionizing the way artificial intelligence can assist in a variety of professional and everyday tasks.
Source: www.theverge.com
